ChatGPT applies limits to conversation context, generated responses, message usage, and uploaded files. The context allowance determines how much information a model can process, and maximum output tokens cap the generated answer. Message and upload allowances govern how often those features can be used.
ChatGPT context allowances vary by plan and thinking level. In Instant, Free has 27K tokens, Go and Plus have 54K, and Pro has 128K.
In the OpenAI API, context and output limits are attached to each model ID. GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna each have a 1,050,000-token context window and a 128,000-token output maximum. GPT-4o has a 128,000-token context window and can generate up to 16,384 tokens.
ChatGPT Free, Go, Plus, and Pro Token Limits
For paid ChatGPT accounts with GPT-6 access, Plus uses GPT-6 Sol for Instant, Medium, and High reasoning. Pro also has Extra High and GPT-6 Pro, which runs on GPT-6 Astra under ChatGPT subscription limits.
GPT-6 Luna began rolling out to Free and Go on October 8, 2026. Accounts awaiting that update use GPT-5.6 Luna for default chat and Think.
| ChatGPT plan | Default Model | Manual reasoning | Context limits |
|---|---|---|---|
| Free | GPT-5.6 Luna GPT-6 Luna rollout | Think button | 27K for Instant Reasoning context varies |
| Go | GPT-5.6 Luna GPT-6 Luna rollout | Think button | 54K for Instant 256K for reasoning |
| Plus | GPT-6 Sol | Medium and High | 54K for Instant 256K for reasoning |
| Pro | GPT-6 Sol | Medium, High, Extra High, and GPT-6 Pro | 128K for Instant 400K for reasoning |
OpenAI API Token Limits by Model
An API request uses the context window and output maximum of the model ID it calls.
For API input, cached-input, and output prices, see the AI LLM API pricing reference.
The January 14, 2027 API shutdown applies to GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini, including their listed dated snapshots and fine-tuned versions. The replacement for these models is gpt-6-luna.
| OpenAI API model | Model ID | Context window | Max output tokens | API status |
|---|---|---|---|---|
| GPT-6 Astra | gpt-6-astra | 1,050,000 | 128,000 | Available |
| GPT-6.1 Sol | gpt-6.1-sol | 1,050,000 | 128,000 | Available |
| GPT-6 Sol | gpt-6-sol | 1,050,000 | 128,000 | Available |
| GPT-6 Luna | gpt-6-luna | 1,050,000 | 128,000 | Available |
| GPT-5.6 Sol | gpt-5.6-solAlias: gpt-5.6 | 1,050,000 | 128,000 | Available |
| GPT-5.6 Terra | gpt-5.6-terra | 1,050,000 | 128,000 | Available |
| GPT-5.6 Luna | gpt-5.6-luna | 1,050,000 | 128,000 | Available |
| GPT-4.1 | gpt-4.1 | 1,047,576 | 32,768 | Deprecated |
| GPT-4.1 mini | gpt-4.1-mini | 1,047,576 | 32,768 | Deprecated |
| GPT-4o | gpt-4o | 128,000 | 16,384 | Deprecated |
| GPT-4o mini | gpt-4o-mini | 128,000 | 16,384 | Deprecated |
GPT-4o Max Output Tokens and Context Window
The gpt-4o API model accepts up to 128,000 tokens of context per request and generates up to 16,384 tokens. The request’s input and generated response both count toward its context window.
A request containing 120,000 input tokens leaves roughly 8,000 tokens for generation in its 128,000-token context window, before overhead. GPT-4o’s 16,384-token output maximum applies only when a request has enough context capacity for a response of that length.
Context Window vs Maximum Output Tokens
A context window is the total token capacity of a model request. It accounts for instructions, input, conversation history supplied with the request, retrieved passages, tool results, and generated tokens.
Maximum output tokens limit what the model can generate. On reasoning models, that includes internal reasoning as well as the visible answer. GPT-4.1, for example, has a 1,047,576-token context window and a 32,768-token output maximum.
The usable response length depends on both limits. Large inputs leave less room for output within the context window, and reasoning tokens consume part of the generation budget. A response can also end earlier when the task is complete, a stop condition is reached, or the request uses a smaller output budget.
Compare OpenAI, Claude, Gemini, and other model limits in the AI context window comparison.
ChatGPT Message Limits Are Different from Token Limits
ChatGPT counts model messages across an account during a defined time period. Starting a new conversation does not reset that allowance. Once it is exhausted, ChatGPT can switch models or suspend access to the selected model until the allowance resets.
Free and Go include unlimited everyday text chats under abuse-prevention safeguards. Think, file uploads, image generation, voice, and data analysis have their own usage allowances for each plan.
GPT-6.1 Sol is available in ChatGPT Work, Codex, and the API, whose usage allowances are distinct from regular Chat. In regular Chat, GPT-6 uses GPT-6 Sol at Instant through Extra High.
GPT-6 Pro Message Limits in ChatGPT
GPT-6 Pro in ChatGPT runs on GPT-6 Astra. Personal access comes with the Pro $100, Pro $200, and Pro $500 subscription tiers. Business and Enterprise workspaces also have access, with Enterprise model availability controlled by workspace administrators.
| Plan | GPT-6 Pro allowance in Chat | Notes |
|---|---|---|
| Pro $500 | Not specified | Highest included usage among Pro tiers. Pro $500 also includes GPT-6 Astra Ultrafast for Work and Codex usage. |
| Pro $200 | 200 messages per week | GPT-5.6 Sol Pro has a 170-message daily allowance. The two Pro models have a combined daily cap of 200 messages. |
| Pro $100 | 50 messages per week | Shared weekly allowance with GPT-5.6 Sol Pro. |
| Business Standard | 15 messages per month | Shared included allowance with GPT-5.6 Sol Pro. |
| Business Premium | 50 messages per week | Shared included allowance with GPT-5.6 Sol Pro. |
ChatGPT File Upload Limits
ChatGPT checks file size, token count, and recent upload usage when you attach a document. During analysis, the selected model processes document content within its context allowance, which can be much smaller than the file’s upload limit.
| File limit | Current value | Scope |
|---|---|---|
| Maximum file size | 512 MB per file | Files uploaded to a GPT or ChatGPT conversation |
| Text and document token cap | 2 million tokens per file | Text and document files (spreadsheets use a file-size cap) |
| Spreadsheet size | Approximately 50 MB | Depends on the size of each row |
| Image size | 20 MB per image | Uploaded images |
| Rolling upload allowance | Up to 80 files every 3 hours | The allowance can be reduced during peak periods |
| Free-plan upload allowance | 3 file uploads per day | ChatGPT Free |
| Stored files | 25 GB per user, 100 GB per organization | Shared across applicable ChatGPT file surfaces |
Token Limits vs OpenAI API Rate Limits
An API model determines the context and output capacity of an individual request. API rate limits measure traffic over time through requests per minute (RPM), tokens per minute (TPM), requests per day (RPD), and batch queue capacity. Account usage tiers can increase those traffic allowances.
See OpenAI API rate limits for current RPM, TPM, daily, and batch-limit details.
What Counts as a Token
Language models process text as tokens, which can represent entire words, word fragments, punctuation, whitespace, numbers, or pieces of code. In ordinary English prose, a token averages roughly four characters or three-quarters of a word.
These averages are less reliable for code, JSON, URLs, logs, and non-English text, which can tokenize differently from English prose. A tokenizer for the selected model gives a more accurate count when a prompt approaches the context limit.
Reasoning Tokens and the API Output Budget
Reasoning models use generated tokens for internal reasoning as well as the visible answer. A request that requires extensive reasoning can use most of its output budget before producing much visible text.
In the Responses API, max_output_tokens sets the budget for reasoning and visible text together. When that budget runs out, the response status becomes incomplete and incomplete_details.reason is max_output_tokens. Response metadata records the total in usage.output_tokens and the reasoning portion in usage.output_tokens_details.reasoning_tokens.
How to Check the Current Limit
- ChatGPT model and mode: Check the model picker for Instant, Medium, High, Extra High, or Pro. To use GPT-5.6 Sol, select it explicitly.
- ChatGPT allowance: Read the in-product notice for the affected model or tool. It can show a reset time, fallback, or temporary unavailability.
- API context and output: Check the exact API model page for the model ID used by the request.
- API throughput: Review the project or organization limits page for RPM, TPM, daily quotas, and batch queue capacity.
- Actual API usage: Inspect response usage metadata for input, output, cached, and reasoning tokens when the selected model reports them.
What to Do When a Limit Is Reached
| Limit reached | Correct action |
|---|---|
| ChatGPT conversation context | Start a clean conversation with a short brief containing the goal, accepted decisions, constraints, and next task |
| ChatGPT message or reasoning allowance | Switch to an available model or mode, or wait for the displayed reset time |
| API context window | Shorten instructions, remove repeated examples, retrieve selected passages, or split the workload across multiple requests |
| API maximum output | Reserve output capacity and request one named section per call |
| ChatGPT file limit | Split large documents, remove unneeded files, delete stored files, or wait for the upload-rate window to reset |
| API rate limit | Reduce concurrency, use retry backoff, queue work, use Batch where appropriate, or move to a higher usage tier when available |
Older and Deprecated OpenAI Model Limits
| Model | Context window | Max output tokens | Status |
|---|---|---|---|
| GPT-4 Turbo | 128,000 | 4,096 | Deprecated (API ends Oct 23, 2026) |
| GPT-4 | 8,192 | 8,192 | Deprecated (API ends Oct 23, 2026) |
| GPT-3.5 Turbo | 16,385 | 4,096 | Deprecated (API ends Oct 23, 2026) |
Real-world Token-Limit Examples
Analyzing a Long PDF in ChatGPT
A PDF can meet ChatGPT’s two-million-token upload allowance and contain more text than the selected model can work with in one response. For a long report, ask about individual chapters or page ranges and retain short summaries of earlier sections when you need to compare findings across the document.
Generating a Long API Report
A report longer than the API model’s maximum output requires multiple requests. Generate sections with clear boundaries, save the responses in your application, and check the assembled report against the outline for consistent terminology and conclusions.
Working with a Large Codebase
For a large-repository debugging task, send the source files related to the failure, relevant dependencies, exact error messages, and a short project map. Use focused log excerpts and omit generated files, vendor directories, or full build logs unless they are relevant to the error.
See Best AI Coding Agents for tools designed around repository workflows.
FAQs
Does ChatGPT Free include GPT-6 now?
GPT-6 Luna began rolling out to Free and Go on October 8, 2026. GPT-5.6 Luna continues to power default chats and Think for accounts that have not received the update.
Is the GPT-6 Astra 1.05M API context window also the ChatGPT GPT-6 Pro token limit?
No. The 1,050,000-token window applies to API requests using gpt-6-astra. GPT-6 Pro conversations follow ChatGPT subscription context limits.
Can I set max_output_tokens above the model’s maximum output?
No. max_output_tokens cannot exceed the output maximum of the model used in the API request.
Does starting a new ChatGPT conversation reset a message allowance?
No. The message allowance is tracked across your account for the relevant time window, including newly created conversations.
Does asking ChatGPT or the API to continue bypass the maximum output limit?
No. Continuing produces another response subject to the model’s output maximum, and any earlier messages supplied again use context capacity.
Should an API application use a model alias or a dated snapshot?
An alias tracks the model version assigned to that ID and can change over time. A dated snapshot fixes a particular version for reproducible behavior until its retirement date.
Related Resources
Changelog
October 20, 2026: Updated the October 23, 2026 and January 14, 2027 API retirement information, including model IDs and replacements. Clarified ChatGPT rollout status, token-budget examples, and FAQ answers.
October 8, 2026: Updated ChatGPT model access after the October 7 GPT-6 announcement. GPT-6 Sol began rolling out on paid Chat plans, and the GPT-6 Luna rollout for Free and Go was scheduled to begin on October 8. Clarified ChatGPT and API context allowances.
September 29, 2026: Added GPT-6.1 Sol API context and output limits.
September 22, 2026: Added GPT-6 Sol and GPT-6 Luna API context and output limits. At that time, Free and Go chat used GPT-5.6 Luna, and GPT-6 Luna was initially available as a desktop-app trial.
September 3, 2026: Added GPT-6 Astra API context and output limits, GPT-6 Pro Chat access and message allowances, and rollout information available at the time.
August 6, 2026: GPT-5.6 Luna began replacing GPT-5.5 Instant as the Free and Go default. The staged rollout introduced unlimited everyday text chats and a Think button for those plans. GPT-5.6 Sol also began rolling out to Plus and Pro with Instant and deeper reasoning options.
August 4, 2026: Updated Free, Plus, and Pro model access and context limits. Added GPT-5.6 reasoning information and refreshed GPT-4o output limits, file-upload limits, reasoning-token behavior, and model deprecations.
July 21, 2026: Updated API limits for GPT-5.6 Sol, Terra, and Luna. Corrected token-limit figures for GPT-4o, GPT-4o mini, and GPT-4.1.
In this article, the “Token Limit in GPT-4” heading, and the “Max tokens” in the table are incorrect. The numbers do not indicate the Max tokens. They are Context Window.
Fixed. I forget to add the ‘Max Output Tokens’ column. Thanks for your feedback.